ATS Job Feeds: Greenhouse, Lever, Ashby, Personio, Workable, Recruitee

Give it company job boards (or just company domains). It reads the public job-board endpoints of Greenhouse, Lever, Ashby, Personio, Workable and Recruitee (official, documented APIs, except Workable's unofficial careers-widget endpoint; Recruitee is read only until 2027-02-10, see below) and returns one clean job schema. Schedule it in monitor mode and it returns only what changed since the last run: new jobs, changed jobs (title, location, salary, remote) and closed jobs with the date they disappeared and how many days they were open (time to fill). It also keeps a per-company posting history, your hiring-velocity signal.

Price: job-returned $0.003 per new, changed, closed (monitor) or open (snapshot) job row; company-checked-unchanged $0.001 when a monitored company has nothing new; board-discovered $0.003 (discover mode only) per company whose board was found and validated. Failed boards, detection misses and store errors are free. Watching 100 companies daily with about 3 changes each a week costs about $7/month (the first check of each company is a baseline at $0.001 unless you turn on emitBaseline; the free baseline_summary row, see "First run", is not charged).

Job data comes only from the ATS endpoints, never from scraped HTML; for a bare company domain the Actor reads at most 3 pages of the company's own site just to find the board link. No search engines, no LinkedIn or Indeed, no logins, no proxies. robots.txt is honoured on every host, with a polite per-host delay.

Input

{
  "companies": [
    "greenhouse:gitlab",
    "https://jobs.lever.co/palantir",
    "https://jobs.ashbyhq.com/ashby",
    "personio.jobs.personio.de",
    "workable:blueground",
    "recruitee:bunq",
    {"domain": "acme.com", "label": "Acme"}
  ],
  "mode": "monitor",
  "keywords": ["engineer", "data"],
  "locations": ["Germany", "Remote"],
  "remoteOnly": false,
  "postedSinceDays": 0
}

Field	What it does
companies	Up to 200. ats:board (greenhouse, lever, lever-eu, ashby, personio, workable, recruitee), a board URL, or a company domain. Workday, SmartRecruiters and Teamtailor entries give a free unsupported_ats error row.
mode	monitor (default): only new / changed / closed jobs since the last run. snapshot: every matching open job. discover: one row per company with the board found for it (see below), no job rows.
keywords, excludeKeywords, keywordsIn	Any-word match in the title (default) or title + description. Case-insensitive.
locations, departments	Any-word match in any location / in the department or team.
remoteOnly	Only jobs marked remote (hybrid is not remote).
postedSinceDays	Only jobs posted in the last N days (0 = off).
maxDescriptionChars	Markdown description cap (default 4,000; 0 = no description).
maxJobsPerCompany	Row (and charge) cap per company per run. In monitor mode the rest come on the next run.
emitBaseline	Monitor: return every open job as new (baseline: true) on a company's first check. Default off: the first check stores the baseline for one company-checked-unchanged ($0.001) and returns one free baseline_summary row. On, 100 companies x 150 jobs = 15,000 rows, about $22.50 once.
includeChanged, includeClosed	Monitor: return changed / closed jobs (default on).
closeAfterMissedChecks	Consecutive checks a job must be missing before it counts as closed (default 2; the closing date is the first missed check).
emitCompanyStats	One free stats row per company (always in the SUMMARY record anyway).
monitorStoreName	Named key-value store with the history (default ats-job-feeds). One schedule per store.

Domain auto-detection. For acme.com the Actor first reads the company's own site: https://acme.com/ and, if that page links to no job board, up to 2 careers pages it links to on the same domain (/careers, /jobs, /join-us, /karriere, ...). Redirects are followed only while they stay on acme.com or its sub-domains; a redirect to another domain ends the site check without contacting that domain. Every supported board URL in those pages counts (links, iframes, Greenhouse embed scripts, URLs inside page JSON); boards whose token matches the domain (acme, acmeinc, acme-labs) are tried first. A linked board is detectionVerified: true only when its token or the API's company name matches the domain (or its job URLs are on the domain); a link to someone else's board (partner, portfolio, customer page) is used but marked false (another company's name) or null. If the site links to no supported board, the Actor guesses the board token (acme; for acme-labs.co.uk also acmelabs) and asks Greenhouse, Lever, Ashby, Personio, Workable and Recruitee once each, in that order (slug-guess): the first board with jobs wins; an existing but empty board is used only if nothing else answers. For guesses detectionVerified is true/false when the API's company name (Greenhouse, Personio, Workable, Recruitee) or job URLs can be compared, null otherwise (Lever, Ashby). At most 3 site pages + 12 guesses per domain. No search engine is ever queried. A site that links only to Workday, SmartRecruiters or Teamtailor gets a free unsupported_ats error naming it. Hits are cached in the monitor store for 30 days, misses for 7 days.

Discover mode ("mode": "discover") runs only that detection for a list of domains and returns one row per company: {status: "discovered", company, domain, ats, board, boardUrl, endpoint, method: "site-link"|"slug-guess"|"given", foundOn, detectionVerified, charged, openJobs}. One board-discovered is charged, after the row is stored, only for boards you gave (given) or boards with detectionVerified: true. Unverified finds are returned free with charged: false, and so are misses (not_detected, unsupported_ats) and errors. Feed the results into monitor mode as ats:board entries.

First run (monitor mode). The first check of a company only stores its open jobs as the baseline, so it returns no job rows. To show that the run worked, it returns one free row with status: "baseline_summary": company, storedJobs (open jobs stored), matchingFilters (how many match your keywords and filters), sample (the first 5 matching jobs: title, location, URL), message, nextStep and charged: false. It has no jobId, is never a charged event and is not a job row; the only charge is the one company-checked-unchanged ($0.001) per company. The run's status message reads "Baseline stored for N of M companies". The second run with no change returns no rows (one company-checked-unchanged per company); changes appear as new, changed and closed. With emitBaseline on there is no baseline_summary row, the jobs themselves are returned.

Filters decide which rows you get and pay for. In monitor mode every job of the company is still tracked, so the history and counts are company-wide.

Output (one row per job)

{
  "status": "closed",
  "jobId": "greenhouse:gitlab:8860302002",
  "ats": "greenhouse", "board": "gitlab", "company": "GitLab",
  "title": "Account Executive - France",
  "location": "Remote, France", "locations": ["Remote, France", "France"],
  "remote": true, "workplaceType": "remote",
  "department": "EMEA - Commercial", "team": null, "employmentType": null,
  "salary": null,
  "postedAt": "2026-10-02T15:31:50Z", "updatedAt": "2026-10-02T15:31:50Z",
  "applyUrl": "https://job-boards.greenhouse.io/gitlab/jobs/8860302002",
  "jobUrl": "https://job-boards.greenhouse.io/gitlab/jobs/8860302002",
  "source": {"ats": "greenhouse", "endpoint": "https://boards-api.greenhouse.io/v1/boards/gitlab/jobs?content=true"},
  "description": "GitLab is the intelligent orchestration platform ...",
  "descriptionTruncated": true, "privacyRedactions": 0,
  "firstSeenAt": "2026-10-14T08:00:00Z", "changes": [],
  "closedAt": "2026-10-24T08:00:00Z", "daysOpen": 22,
  "checkedAt": "2026-10-24T08:00:00Z", "error": null
}

• status: baseline_summary (free, first monitor check only, see "First run"), new, changed (with changes: [{field, before, after}] for title, locations, salary, remote), closed, open (snapshot), error (free), company (free stats row).
• remote: true/false, or null when the board does not say. salary is filled only when the board publishes it: Lever salaryRange, Ashby compensation, and Greenhouse pay ranges rendered into the post. Personio feeds carry no salary.
• Closed rows carry what the monitor remembered (title, locations, department, salary text, posted / first-seen date). daysOpen counts from the posting date (else the first time the Actor saw the job) to closedAt, the first check where the job was missing.
• SUMMARY record: per company openJobs, new30d / closed30d / new90d / closed90d, netChange30d / netChange90d, medianDaysOpenOfClosed, trackedSince, plus counts and warnings.

Run-summary row (free). When a run produces no other row (for example a monitor run where nothing changed, or a run stopped at once by your maximum charge or the run timeout), the Actor adds one row {"type": "run_summary", "status": "run_summary", "message": "...", "stopReason": ..., "warnings": [...], "summary": {...}} (summary is the same as the SUMMARY record), so the dataset is never empty. It is never charged. Skip it with type != "run_summary".

Privacy: no personal data

• ATS fields that name people (Greenhouse custom metadata such as "Hiring Manager", hiring-team data) are never read.
• In descriptions, departments, locations and salary text: e-mail addresses (also name (at) domain (dot) com), phone numbers (international, US incl. space-only and 1-800, German 089 / 123 456 78), LinkedIn / Calendly / WhatsApp / Telegram profile links and mailto/tel links are removed. So are names after a cue: strong cues ("I'm", "my name is", "Recruiter:", "Hiring Manager -", "Ansprechpartner(in):") take one capitalised word; weak cues ("contact", "reach out to", "report to") need two words or an honorific (Dr., Mr., Ms., Frau, Herr). Hyphenated surnames and a comma after the cue are handled. Words like products and companies ("contact Google Cloud") and roles are never treated as names. Job titles only lose e-mails, phones and links, never words. Each removal is replaced by a marker such as [email removed] and counted in privacyRedactions.
• Limit: this is a heuristic. A name with no cue ("Abhik and Benji, our founders") is not detected. Founders and executives named in a company's public job text may remain.

Monitor mode, how it works

Each company has its own records in the named key-value store: state-<ats>-<board> (one compact entry per open job: title, locations, salary text, remote, department and dates, never descriptions), history-<ats>-<board> (daily open/new/closed/changed counts for 400 days, plus the last 500 closed jobs with days open), and a LOCK while a run is active.

• A job counts as closed after it has been missing from closeAfterMissedChecks (default 2) consecutive successful answers. ATS APIs can briefly serve stale or partial lists, so a single missing check is treated as possibly a glitch. A board that fails to load (timeout, 5xx, robots) closes nothing. Whatever the setting, at least two checks are needed when a board goes empty (board_empty_unconfirmed), shrinks by 30 % or more and at least 5 jobs (board_shrink_unconfirmed), or is a multi-page Lever board (multi_page_board). closedAt is still the first missed check. This lowers false closures but cannot rule them out.
• A job that disappears and comes back within closeAfterMissedChecks is neither new nor closed.
• Billing order: the state record is saved first, then the rows are pushed and charged. If the state save fails, nothing is pushed or charged and the same changes come next run. If the history record fails to save twice in a row, that company is checked for free until a save works again (SUMMARY.chargingPaused).
• The state size is checked before any charge (1.5 MB cap, state_too_large, free).
• When your maximum charge or maxJobsPerCompany cuts a company short, the rows not delivered are rolled back in the state (a new job is "not seen yet", a closed job keeps its closing date) and come on the next run. Nothing is lost or charged twice (deferred in SUMMARY). A company is not started at all when not even one row fits.
• Use one schedule per monitorStoreName. A second run on the same store while one is active stops with store_locked (a crashed run's lock expires after 3 hours). The lock is best effort: the key-value store has no atomic compare-and-set, so two runs started in the same second can both proceed. A run removes the lock only if it is still its own.
• If a delivery is interrupted (the platform refuses a row), the undelivered rows are rolled back and the state is saved again, so they come next run. Baseline jobs deferred by the row cap stay marked baseline: true and are never counted as new hires in the history.

Sources and their terms

Public endpoints that need no login or key: Greenhouse, Lever, Ashby, Personio and Recruitee (official, documented; Recruitee only until 2027-02-10) and one unofficial endpoint (Workable's careers widget). All six are enabled by default; nothing needs switching on. Details, terms, robots.txt and rate-limit notes are in SOURCES.md.

ATS	Endpoint	Notes
Greenhouse	boards-api.greenhouse.io/v1/boards/{board}/jobs?content=true	"Job Board data is publicly available"; one request per board
Lever	api.lever.co/v0/postings/{board}?mode=json (EU: api.eu.lever.co)	published postings "may be scraped by third parties"; 100 per page; robots Crawl-delay 1
Ashby	api.ashbyhq.com/posting-api/job-board/{board}?includeCompensation=true	public posting API for job boards
Personio	{board}.jobs.personio.de/xml	documented open-positions XML feed; parsed with defusedxml
Workable	apply.workable.com/api/v1/widget/accounts/{board}?details=true	unofficial widget endpoint: the careers-widget endpoint Workable gives customers to embed jobs, no formal API reference. Workable's terms have no clause against automated access; robots.txt allows all
Recruitee	{board}.recruitee.com/api/offers/	public Careers Site API; the job's reply-to mailbox, screening questions and custom fields are never read

Live check on 2026-10-14 (one request per endpoint, robots.txt first, User-Agent DataFeedBot): Workable (Blueground) and Recruitee (bunq) answered public JSON without login. Not included: Workday (*.myworkdayjobs.com). Its JSON list works without login, but Workday's site terms (https://workday.com/en-us/legal/site-terms.html, section 2) ban "robots or similar data gathering or extraction methods designed to scrape" and their "Sites" include "the Workday APIs", so it is disabled. The adapter stays in the code, switched off. SmartRecruiters: robots.txt of api.smartrecruiters.com allows only LinkedInBot. Teamtailor: its JSON API needs the company's own key. iCIMS, Taleo, SuccessFactors: no public JSON. Recruitee: Recruitee announced that from 2027-02-10 the Careers Site API needs a company token. From that date the Actor stops calling it (checked in the code before any request) and Recruitee boards get a free unsupported_ats error row; domain auto-detection no longer finds Recruitee boards either. While you still list Recruitee boards, SUMMARY.warnings says how many days remain, and after the date that the platform is no longer read.

Limits

• Board responses over 10 MB are refused (too_large). That is roughly 1,000+ Greenhouse jobs with full descriptions; maxDescriptionChars does not reduce the download. Lever is paged and allows up to 5,000 postings. A board whose monitor state exceeds 1.5 MB (several thousand jobs) gets a free state_too_large row: run that board in snapshot mode instead.
• Greenhouse EU boards (job-boards.eu.greenhouse.io) are not supported yet.
• Slug guessing uses the domain name only; site-link detection reads at most 3 pages of the company's own site (robots.txt honoured) and does not run JavaScript, so boards injected only by script at runtime are missed.
• Recruitee and Workable boards on a company's own career domain (e.g. careers.acme.com served by Recruitee) are not recognised from links; give recruitee:acme instead.
• Salaries are as published. No currency conversion, no salary parsing from free text (except Greenhouse's pay-range block, which understands US and EU separators, K suffixes and C$/A$/S$/NZ$). When an amount is ambiguous, min/max stay null and only text is filled.
• employmentType is not exposed by the Greenhouse Job Board API (null).

Acceptable use

The Actor reads job postings that companies publish for anyone to read through their ATS's public API. It outputs company-level job data only, never applicant or recruiter data. You are responsible for how you use the data, including the ATS terms and applicable law such as the GDPR.